xref: /openbmc/linux/arch/x86/kernel/sev.c (revision 413d6ed3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * AMD Memory Encryption Support
4  *
5  * Copyright (C) 2019 SUSE
6  *
7  * Author: Joerg Roedel <jroedel@suse.de>
8  */
9 
10 #define pr_fmt(fmt)	"SEV-ES: " fmt
11 
12 #include <linux/sched/debug.h>	/* For show_regs() */
13 #include <linux/percpu-defs.h>
14 #include <linux/mem_encrypt.h>
15 #include <linux/lockdep.h>
16 #include <linux/printk.h>
17 #include <linux/mm_types.h>
18 #include <linux/set_memory.h>
19 #include <linux/memblock.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 
23 #include <asm/cpu_entry_area.h>
24 #include <asm/stacktrace.h>
25 #include <asm/sev.h>
26 #include <asm/insn-eval.h>
27 #include <asm/fpu/internal.h>
28 #include <asm/processor.h>
29 #include <asm/realmode.h>
30 #include <asm/traps.h>
31 #include <asm/svm.h>
32 #include <asm/smp.h>
33 #include <asm/cpu.h>
34 
35 #define DR7_RESET_VALUE        0x400
36 
37 /* For early boot hypervisor communication in SEV-ES enabled guests */
38 static struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);
39 
40 /*
41  * Needs to be in the .data section because we need it NULL before bss is
42  * cleared
43  */
44 static struct ghcb __initdata *boot_ghcb;
45 
46 /* #VC handler runtime per-CPU data */
47 struct sev_es_runtime_data {
48 	struct ghcb ghcb_page;
49 
50 	/* Physical storage for the per-CPU IST stack of the #VC handler */
51 	char ist_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
52 
53 	/*
54 	 * Physical storage for the per-CPU fall-back stack of the #VC handler.
55 	 * The fall-back stack is used when it is not safe to switch back to the
56 	 * interrupted stack in the #VC entry code.
57 	 */
58 	char fallback_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
59 
60 	/*
61 	 * Reserve one page per CPU as backup storage for the unencrypted GHCB.
62 	 * It is needed when an NMI happens while the #VC handler uses the real
63 	 * GHCB, and the NMI handler itself is causing another #VC exception. In
64 	 * that case the GHCB content of the first handler needs to be backed up
65 	 * and restored.
66 	 */
67 	struct ghcb backup_ghcb;
68 
69 	/*
70 	 * Mark the per-cpu GHCBs as in-use to detect nested #VC exceptions.
71 	 * There is no need for it to be atomic, because nothing is written to
72 	 * the GHCB between the read and the write of ghcb_active. So it is safe
73 	 * to use it when a nested #VC exception happens before the write.
74 	 *
75 	 * This is necessary for example in the #VC->NMI->#VC case when the NMI
76 	 * happens while the first #VC handler uses the GHCB. When the NMI code
77 	 * raises a second #VC handler it might overwrite the contents of the
78 	 * GHCB written by the first handler. To avoid this the content of the
79 	 * GHCB is saved and restored when the GHCB is detected to be in use
80 	 * already.
81 	 */
82 	bool ghcb_active;
83 	bool backup_ghcb_active;
84 
85 	/*
86 	 * Cached DR7 value - write it on DR7 writes and return it on reads.
87 	 * That value will never make it to the real hardware DR7 as debugging
88 	 * is currently unsupported in SEV-ES guests.
89 	 */
90 	unsigned long dr7;
91 };
92 
93 struct ghcb_state {
94 	struct ghcb *ghcb;
95 };
96 
97 static DEFINE_PER_CPU(struct sev_es_runtime_data*, runtime_data);
98 DEFINE_STATIC_KEY_FALSE(sev_es_enable_key);
99 
100 /* Needed in vc_early_forward_exception */
101 void do_early_exception(struct pt_regs *regs, int trapnr);
102 
103 static void __init setup_vc_stacks(int cpu)
104 {
105 	struct sev_es_runtime_data *data;
106 	struct cpu_entry_area *cea;
107 	unsigned long vaddr;
108 	phys_addr_t pa;
109 
110 	data = per_cpu(runtime_data, cpu);
111 	cea  = get_cpu_entry_area(cpu);
112 
113 	/* Map #VC IST stack */
114 	vaddr = CEA_ESTACK_BOT(&cea->estacks, VC);
115 	pa    = __pa(data->ist_stack);
116 	cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
117 
118 	/* Map VC fall-back stack */
119 	vaddr = CEA_ESTACK_BOT(&cea->estacks, VC2);
120 	pa    = __pa(data->fallback_stack);
121 	cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
122 }
123 
124 static __always_inline bool on_vc_stack(struct pt_regs *regs)
125 {
126 	unsigned long sp = regs->sp;
127 
128 	/* User-mode RSP is not trusted */
129 	if (user_mode(regs))
130 		return false;
131 
132 	/* SYSCALL gap still has user-mode RSP */
133 	if (ip_within_syscall_gap(regs))
134 		return false;
135 
136 	return ((sp >= __this_cpu_ist_bottom_va(VC)) && (sp < __this_cpu_ist_top_va(VC)));
137 }
138 
139 /*
140  * This function handles the case when an NMI is raised in the #VC
141  * exception handler entry code, before the #VC handler has switched off
142  * its IST stack. In this case, the IST entry for #VC must be adjusted,
143  * so that any nested #VC exception will not overwrite the stack
144  * contents of the interrupted #VC handler.
145  *
146  * The IST entry is adjusted unconditionally so that it can be also be
147  * unconditionally adjusted back in __sev_es_ist_exit(). Otherwise a
148  * nested sev_es_ist_exit() call may adjust back the IST entry too
149  * early.
150  *
151  * The __sev_es_ist_enter() and __sev_es_ist_exit() functions always run
152  * on the NMI IST stack, as they are only called from NMI handling code
153  * right now.
154  */
155 void noinstr __sev_es_ist_enter(struct pt_regs *regs)
156 {
157 	unsigned long old_ist, new_ist;
158 
159 	/* Read old IST entry */
160 	new_ist = old_ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
161 
162 	/*
163 	 * If NMI happened while on the #VC IST stack, set the new IST
164 	 * value below regs->sp, so that the interrupted stack frame is
165 	 * not overwritten by subsequent #VC exceptions.
166 	 */
167 	if (on_vc_stack(regs))
168 		new_ist = regs->sp;
169 
170 	/*
171 	 * Reserve additional 8 bytes and store old IST value so this
172 	 * adjustment can be unrolled in __sev_es_ist_exit().
173 	 */
174 	new_ist -= sizeof(old_ist);
175 	*(unsigned long *)new_ist = old_ist;
176 
177 	/* Set new IST entry */
178 	this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], new_ist);
179 }
180 
181 void noinstr __sev_es_ist_exit(void)
182 {
183 	unsigned long ist;
184 
185 	/* Read IST entry */
186 	ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
187 
188 	if (WARN_ON(ist == __this_cpu_ist_top_va(VC)))
189 		return;
190 
191 	/* Read back old IST entry and write it to the TSS */
192 	this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], *(unsigned long *)ist);
193 }
194 
195 static __always_inline struct ghcb *sev_es_get_ghcb(struct ghcb_state *state)
196 {
197 	struct sev_es_runtime_data *data;
198 	struct ghcb *ghcb;
199 
200 	data = this_cpu_read(runtime_data);
201 	ghcb = &data->ghcb_page;
202 
203 	if (unlikely(data->ghcb_active)) {
204 		/* GHCB is already in use - save its contents */
205 
206 		if (unlikely(data->backup_ghcb_active))
207 			return NULL;
208 
209 		/* Mark backup_ghcb active before writing to it */
210 		data->backup_ghcb_active = true;
211 
212 		state->ghcb = &data->backup_ghcb;
213 
214 		/* Backup GHCB content */
215 		*state->ghcb = *ghcb;
216 	} else {
217 		state->ghcb = NULL;
218 		data->ghcb_active = true;
219 	}
220 
221 	return ghcb;
222 }
223 
224 static __always_inline void sev_es_put_ghcb(struct ghcb_state *state)
225 {
226 	struct sev_es_runtime_data *data;
227 	struct ghcb *ghcb;
228 
229 	data = this_cpu_read(runtime_data);
230 	ghcb = &data->ghcb_page;
231 
232 	if (state->ghcb) {
233 		/* Restore GHCB from Backup */
234 		*ghcb = *state->ghcb;
235 		data->backup_ghcb_active = false;
236 		state->ghcb = NULL;
237 	} else {
238 		data->ghcb_active = false;
239 	}
240 }
241 
242 /* Needed in vc_early_forward_exception */
243 void do_early_exception(struct pt_regs *regs, int trapnr);
244 
245 static inline u64 sev_es_rd_ghcb_msr(void)
246 {
247 	return __rdmsr(MSR_AMD64_SEV_ES_GHCB);
248 }
249 
250 static __always_inline void sev_es_wr_ghcb_msr(u64 val)
251 {
252 	u32 low, high;
253 
254 	low  = (u32)(val);
255 	high = (u32)(val >> 32);
256 
257 	native_wrmsr(MSR_AMD64_SEV_ES_GHCB, low, high);
258 }
259 
260 static int vc_fetch_insn_kernel(struct es_em_ctxt *ctxt,
261 				unsigned char *buffer)
262 {
263 	return copy_from_kernel_nofault(buffer, (unsigned char *)ctxt->regs->ip, MAX_INSN_SIZE);
264 }
265 
266 static enum es_result __vc_decode_user_insn(struct es_em_ctxt *ctxt)
267 {
268 	char buffer[MAX_INSN_SIZE];
269 	int res;
270 
271 	res = insn_fetch_from_user_inatomic(ctxt->regs, buffer);
272 	if (!res) {
273 		ctxt->fi.vector     = X86_TRAP_PF;
274 		ctxt->fi.error_code = X86_PF_INSTR | X86_PF_USER;
275 		ctxt->fi.cr2        = ctxt->regs->ip;
276 		return ES_EXCEPTION;
277 	}
278 
279 	if (!insn_decode_from_regs(&ctxt->insn, ctxt->regs, buffer, res))
280 		return ES_DECODE_FAILED;
281 
282 	if (ctxt->insn.immediate.got)
283 		return ES_OK;
284 	else
285 		return ES_DECODE_FAILED;
286 }
287 
288 static enum es_result __vc_decode_kern_insn(struct es_em_ctxt *ctxt)
289 {
290 	char buffer[MAX_INSN_SIZE];
291 	int res, ret;
292 
293 	res = vc_fetch_insn_kernel(ctxt, buffer);
294 	if (res) {
295 		ctxt->fi.vector     = X86_TRAP_PF;
296 		ctxt->fi.error_code = X86_PF_INSTR;
297 		ctxt->fi.cr2        = ctxt->regs->ip;
298 		return ES_EXCEPTION;
299 	}
300 
301 	ret = insn_decode(&ctxt->insn, buffer, MAX_INSN_SIZE, INSN_MODE_64);
302 	if (ret < 0)
303 		return ES_DECODE_FAILED;
304 	else
305 		return ES_OK;
306 }
307 
308 static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)
309 {
310 	if (user_mode(ctxt->regs))
311 		return __vc_decode_user_insn(ctxt);
312 	else
313 		return __vc_decode_kern_insn(ctxt);
314 }
315 
316 static enum es_result vc_write_mem(struct es_em_ctxt *ctxt,
317 				   char *dst, char *buf, size_t size)
318 {
319 	unsigned long error_code = X86_PF_PROT | X86_PF_WRITE;
320 	char __user *target = (char __user *)dst;
321 	u64 d8;
322 	u32 d4;
323 	u16 d2;
324 	u8  d1;
325 
326 	/* If instruction ran in kernel mode and the I/O buffer is in kernel space */
327 	if (!user_mode(ctxt->regs) && !access_ok(target, size)) {
328 		memcpy(dst, buf, size);
329 		return ES_OK;
330 	}
331 
332 	switch (size) {
333 	case 1:
334 		memcpy(&d1, buf, 1);
335 		if (put_user(d1, target))
336 			goto fault;
337 		break;
338 	case 2:
339 		memcpy(&d2, buf, 2);
340 		if (put_user(d2, target))
341 			goto fault;
342 		break;
343 	case 4:
344 		memcpy(&d4, buf, 4);
345 		if (put_user(d4, target))
346 			goto fault;
347 		break;
348 	case 8:
349 		memcpy(&d8, buf, 8);
350 		if (put_user(d8, target))
351 			goto fault;
352 		break;
353 	default:
354 		WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
355 		return ES_UNSUPPORTED;
356 	}
357 
358 	return ES_OK;
359 
360 fault:
361 	if (user_mode(ctxt->regs))
362 		error_code |= X86_PF_USER;
363 
364 	ctxt->fi.vector = X86_TRAP_PF;
365 	ctxt->fi.error_code = error_code;
366 	ctxt->fi.cr2 = (unsigned long)dst;
367 
368 	return ES_EXCEPTION;
369 }
370 
371 static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,
372 				  char *src, char *buf, size_t size)
373 {
374 	unsigned long error_code = X86_PF_PROT;
375 	char __user *s = (char __user *)src;
376 	u64 d8;
377 	u32 d4;
378 	u16 d2;
379 	u8  d1;
380 
381 	/* If instruction ran in kernel mode and the I/O buffer is in kernel space */
382 	if (!user_mode(ctxt->regs) && !access_ok(s, size)) {
383 		memcpy(buf, src, size);
384 		return ES_OK;
385 	}
386 
387 	switch (size) {
388 	case 1:
389 		if (get_user(d1, s))
390 			goto fault;
391 		memcpy(buf, &d1, 1);
392 		break;
393 	case 2:
394 		if (get_user(d2, s))
395 			goto fault;
396 		memcpy(buf, &d2, 2);
397 		break;
398 	case 4:
399 		if (get_user(d4, s))
400 			goto fault;
401 		memcpy(buf, &d4, 4);
402 		break;
403 	case 8:
404 		if (get_user(d8, s))
405 			goto fault;
406 		memcpy(buf, &d8, 8);
407 		break;
408 	default:
409 		WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
410 		return ES_UNSUPPORTED;
411 	}
412 
413 	return ES_OK;
414 
415 fault:
416 	if (user_mode(ctxt->regs))
417 		error_code |= X86_PF_USER;
418 
419 	ctxt->fi.vector = X86_TRAP_PF;
420 	ctxt->fi.error_code = error_code;
421 	ctxt->fi.cr2 = (unsigned long)src;
422 
423 	return ES_EXCEPTION;
424 }
425 
426 static enum es_result vc_slow_virt_to_phys(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
427 					   unsigned long vaddr, phys_addr_t *paddr)
428 {
429 	unsigned long va = (unsigned long)vaddr;
430 	unsigned int level;
431 	phys_addr_t pa;
432 	pgd_t *pgd;
433 	pte_t *pte;
434 
435 	pgd = __va(read_cr3_pa());
436 	pgd = &pgd[pgd_index(va)];
437 	pte = lookup_address_in_pgd(pgd, va, &level);
438 	if (!pte) {
439 		ctxt->fi.vector     = X86_TRAP_PF;
440 		ctxt->fi.cr2        = vaddr;
441 		ctxt->fi.error_code = 0;
442 
443 		if (user_mode(ctxt->regs))
444 			ctxt->fi.error_code |= X86_PF_USER;
445 
446 		return ES_EXCEPTION;
447 	}
448 
449 	if (WARN_ON_ONCE(pte_val(*pte) & _PAGE_ENC))
450 		/* Emulated MMIO to/from encrypted memory not supported */
451 		return ES_UNSUPPORTED;
452 
453 	pa = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
454 	pa |= va & ~page_level_mask(level);
455 
456 	*paddr = pa;
457 
458 	return ES_OK;
459 }
460 
461 /* Include code shared with pre-decompression boot stage */
462 #include "sev-shared.c"
463 
464 void noinstr __sev_es_nmi_complete(void)
465 {
466 	struct ghcb_state state;
467 	struct ghcb *ghcb;
468 
469 	ghcb = sev_es_get_ghcb(&state);
470 
471 	vc_ghcb_invalidate(ghcb);
472 	ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_NMI_COMPLETE);
473 	ghcb_set_sw_exit_info_1(ghcb, 0);
474 	ghcb_set_sw_exit_info_2(ghcb, 0);
475 
476 	sev_es_wr_ghcb_msr(__pa_nodebug(ghcb));
477 	VMGEXIT();
478 
479 	sev_es_put_ghcb(&state);
480 }
481 
482 static u64 get_jump_table_addr(void)
483 {
484 	struct ghcb_state state;
485 	unsigned long flags;
486 	struct ghcb *ghcb;
487 	u64 ret = 0;
488 
489 	local_irq_save(flags);
490 
491 	ghcb = sev_es_get_ghcb(&state);
492 
493 	vc_ghcb_invalidate(ghcb);
494 	ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_JUMP_TABLE);
495 	ghcb_set_sw_exit_info_1(ghcb, SVM_VMGEXIT_GET_AP_JUMP_TABLE);
496 	ghcb_set_sw_exit_info_2(ghcb, 0);
497 
498 	sev_es_wr_ghcb_msr(__pa(ghcb));
499 	VMGEXIT();
500 
501 	if (ghcb_sw_exit_info_1_is_valid(ghcb) &&
502 	    ghcb_sw_exit_info_2_is_valid(ghcb))
503 		ret = ghcb->save.sw_exit_info_2;
504 
505 	sev_es_put_ghcb(&state);
506 
507 	local_irq_restore(flags);
508 
509 	return ret;
510 }
511 
512 int sev_es_setup_ap_jump_table(struct real_mode_header *rmh)
513 {
514 	u16 startup_cs, startup_ip;
515 	phys_addr_t jump_table_pa;
516 	u64 jump_table_addr;
517 	u16 __iomem *jump_table;
518 
519 	jump_table_addr = get_jump_table_addr();
520 
521 	/* On UP guests there is no jump table so this is not a failure */
522 	if (!jump_table_addr)
523 		return 0;
524 
525 	/* Check if AP Jump Table is page-aligned */
526 	if (jump_table_addr & ~PAGE_MASK)
527 		return -EINVAL;
528 
529 	jump_table_pa = jump_table_addr & PAGE_MASK;
530 
531 	startup_cs = (u16)(rmh->trampoline_start >> 4);
532 	startup_ip = (u16)(rmh->sev_es_trampoline_start -
533 			   rmh->trampoline_start);
534 
535 	jump_table = ioremap_encrypted(jump_table_pa, PAGE_SIZE);
536 	if (!jump_table)
537 		return -EIO;
538 
539 	writew(startup_ip, &jump_table[0]);
540 	writew(startup_cs, &jump_table[1]);
541 
542 	iounmap(jump_table);
543 
544 	return 0;
545 }
546 
547 /*
548  * This is needed by the OVMF UEFI firmware which will use whatever it finds in
549  * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu
550  * runtime GHCBs used by the kernel are also mapped in the EFI page-table.
551  */
552 int __init sev_es_efi_map_ghcbs(pgd_t *pgd)
553 {
554 	struct sev_es_runtime_data *data;
555 	unsigned long address, pflags;
556 	int cpu;
557 	u64 pfn;
558 
559 	if (!sev_es_active())
560 		return 0;
561 
562 	pflags = _PAGE_NX | _PAGE_RW;
563 
564 	for_each_possible_cpu(cpu) {
565 		data = per_cpu(runtime_data, cpu);
566 
567 		address = __pa(&data->ghcb_page);
568 		pfn = address >> PAGE_SHIFT;
569 
570 		if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags))
571 			return 1;
572 	}
573 
574 	return 0;
575 }
576 
577 static enum es_result vc_handle_msr(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
578 {
579 	struct pt_regs *regs = ctxt->regs;
580 	enum es_result ret;
581 	u64 exit_info_1;
582 
583 	/* Is it a WRMSR? */
584 	exit_info_1 = (ctxt->insn.opcode.bytes[1] == 0x30) ? 1 : 0;
585 
586 	ghcb_set_rcx(ghcb, regs->cx);
587 	if (exit_info_1) {
588 		ghcb_set_rax(ghcb, regs->ax);
589 		ghcb_set_rdx(ghcb, regs->dx);
590 	}
591 
592 	ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_MSR, exit_info_1, 0);
593 
594 	if ((ret == ES_OK) && (!exit_info_1)) {
595 		regs->ax = ghcb->save.rax;
596 		regs->dx = ghcb->save.rdx;
597 	}
598 
599 	return ret;
600 }
601 
602 /*
603  * This function runs on the first #VC exception after the kernel
604  * switched to virtual addresses.
605  */
606 static bool __init sev_es_setup_ghcb(void)
607 {
608 	/* First make sure the hypervisor talks a supported protocol. */
609 	if (!sev_es_negotiate_protocol())
610 		return false;
611 
612 	/*
613 	 * Clear the boot_ghcb. The first exception comes in before the bss
614 	 * section is cleared.
615 	 */
616 	memset(&boot_ghcb_page, 0, PAGE_SIZE);
617 
618 	/* Alright - Make the boot-ghcb public */
619 	boot_ghcb = &boot_ghcb_page;
620 
621 	return true;
622 }
623 
624 #ifdef CONFIG_HOTPLUG_CPU
625 static void sev_es_ap_hlt_loop(void)
626 {
627 	struct ghcb_state state;
628 	struct ghcb *ghcb;
629 
630 	ghcb = sev_es_get_ghcb(&state);
631 
632 	while (true) {
633 		vc_ghcb_invalidate(ghcb);
634 		ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_HLT_LOOP);
635 		ghcb_set_sw_exit_info_1(ghcb, 0);
636 		ghcb_set_sw_exit_info_2(ghcb, 0);
637 
638 		sev_es_wr_ghcb_msr(__pa(ghcb));
639 		VMGEXIT();
640 
641 		/* Wakeup signal? */
642 		if (ghcb_sw_exit_info_2_is_valid(ghcb) &&
643 		    ghcb->save.sw_exit_info_2)
644 			break;
645 	}
646 
647 	sev_es_put_ghcb(&state);
648 }
649 
650 /*
651  * Play_dead handler when running under SEV-ES. This is needed because
652  * the hypervisor can't deliver an SIPI request to restart the AP.
653  * Instead the kernel has to issue a VMGEXIT to halt the VCPU until the
654  * hypervisor wakes it up again.
655  */
656 static void sev_es_play_dead(void)
657 {
658 	play_dead_common();
659 
660 	/* IRQs now disabled */
661 
662 	sev_es_ap_hlt_loop();
663 
664 	/*
665 	 * If we get here, the VCPU was woken up again. Jump to CPU
666 	 * startup code to get it back online.
667 	 */
668 	start_cpu0();
669 }
670 #else  /* CONFIG_HOTPLUG_CPU */
671 #define sev_es_play_dead	native_play_dead
672 #endif /* CONFIG_HOTPLUG_CPU */
673 
674 #ifdef CONFIG_SMP
675 static void __init sev_es_setup_play_dead(void)
676 {
677 	smp_ops.play_dead = sev_es_play_dead;
678 }
679 #else
680 static inline void sev_es_setup_play_dead(void) { }
681 #endif
682 
683 static void __init alloc_runtime_data(int cpu)
684 {
685 	struct sev_es_runtime_data *data;
686 
687 	data = memblock_alloc(sizeof(*data), PAGE_SIZE);
688 	if (!data)
689 		panic("Can't allocate SEV-ES runtime data");
690 
691 	per_cpu(runtime_data, cpu) = data;
692 }
693 
694 static void __init init_ghcb(int cpu)
695 {
696 	struct sev_es_runtime_data *data;
697 	int err;
698 
699 	data = per_cpu(runtime_data, cpu);
700 
701 	err = early_set_memory_decrypted((unsigned long)&data->ghcb_page,
702 					 sizeof(data->ghcb_page));
703 	if (err)
704 		panic("Can't map GHCBs unencrypted");
705 
706 	memset(&data->ghcb_page, 0, sizeof(data->ghcb_page));
707 
708 	data->ghcb_active = false;
709 	data->backup_ghcb_active = false;
710 }
711 
712 void __init sev_es_init_vc_handling(void)
713 {
714 	int cpu;
715 
716 	BUILD_BUG_ON(offsetof(struct sev_es_runtime_data, ghcb_page) % PAGE_SIZE);
717 
718 	if (!sev_es_active())
719 		return;
720 
721 	if (!sev_es_check_cpu_features())
722 		panic("SEV-ES CPU Features missing");
723 
724 	/* Enable SEV-ES special handling */
725 	static_branch_enable(&sev_es_enable_key);
726 
727 	/* Initialize per-cpu GHCB pages */
728 	for_each_possible_cpu(cpu) {
729 		alloc_runtime_data(cpu);
730 		init_ghcb(cpu);
731 		setup_vc_stacks(cpu);
732 	}
733 
734 	sev_es_setup_play_dead();
735 
736 	/* Secondary CPUs use the runtime #VC handler */
737 	initial_vc_handler = (unsigned long)safe_stack_exc_vmm_communication;
738 }
739 
740 static void __init vc_early_forward_exception(struct es_em_ctxt *ctxt)
741 {
742 	int trapnr = ctxt->fi.vector;
743 
744 	if (trapnr == X86_TRAP_PF)
745 		native_write_cr2(ctxt->fi.cr2);
746 
747 	ctxt->regs->orig_ax = ctxt->fi.error_code;
748 	do_early_exception(ctxt->regs, trapnr);
749 }
750 
751 static long *vc_insn_get_reg(struct es_em_ctxt *ctxt)
752 {
753 	long *reg_array;
754 	int offset;
755 
756 	reg_array = (long *)ctxt->regs;
757 	offset    = insn_get_modrm_reg_off(&ctxt->insn, ctxt->regs);
758 
759 	if (offset < 0)
760 		return NULL;
761 
762 	offset /= sizeof(long);
763 
764 	return reg_array + offset;
765 }
766 
767 static long *vc_insn_get_rm(struct es_em_ctxt *ctxt)
768 {
769 	long *reg_array;
770 	int offset;
771 
772 	reg_array = (long *)ctxt->regs;
773 	offset    = insn_get_modrm_rm_off(&ctxt->insn, ctxt->regs);
774 
775 	if (offset < 0)
776 		return NULL;
777 
778 	offset /= sizeof(long);
779 
780 	return reg_array + offset;
781 }
782 static enum es_result vc_do_mmio(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
783 				 unsigned int bytes, bool read)
784 {
785 	u64 exit_code, exit_info_1, exit_info_2;
786 	unsigned long ghcb_pa = __pa(ghcb);
787 	enum es_result res;
788 	phys_addr_t paddr;
789 	void __user *ref;
790 
791 	ref = insn_get_addr_ref(&ctxt->insn, ctxt->regs);
792 	if (ref == (void __user *)-1L)
793 		return ES_UNSUPPORTED;
794 
795 	exit_code = read ? SVM_VMGEXIT_MMIO_READ : SVM_VMGEXIT_MMIO_WRITE;
796 
797 	res = vc_slow_virt_to_phys(ghcb, ctxt, (unsigned long)ref, &paddr);
798 	if (res != ES_OK) {
799 		if (res == ES_EXCEPTION && !read)
800 			ctxt->fi.error_code |= X86_PF_WRITE;
801 
802 		return res;
803 	}
804 
805 	exit_info_1 = paddr;
806 	/* Can never be greater than 8 */
807 	exit_info_2 = bytes;
808 
809 	ghcb_set_sw_scratch(ghcb, ghcb_pa + offsetof(struct ghcb, shared_buffer));
810 
811 	return sev_es_ghcb_hv_call(ghcb, ctxt, exit_code, exit_info_1, exit_info_2);
812 }
813 
814 static enum es_result vc_handle_mmio_twobyte_ops(struct ghcb *ghcb,
815 						 struct es_em_ctxt *ctxt)
816 {
817 	struct insn *insn = &ctxt->insn;
818 	unsigned int bytes = 0;
819 	enum es_result ret;
820 	int sign_byte;
821 	long *reg_data;
822 
823 	switch (insn->opcode.bytes[1]) {
824 		/* MMIO Read w/ zero-extension */
825 	case 0xb6:
826 		bytes = 1;
827 		fallthrough;
828 	case 0xb7:
829 		if (!bytes)
830 			bytes = 2;
831 
832 		ret = vc_do_mmio(ghcb, ctxt, bytes, true);
833 		if (ret)
834 			break;
835 
836 		/* Zero extend based on operand size */
837 		reg_data = vc_insn_get_reg(ctxt);
838 		if (!reg_data)
839 			return ES_DECODE_FAILED;
840 
841 		memset(reg_data, 0, insn->opnd_bytes);
842 
843 		memcpy(reg_data, ghcb->shared_buffer, bytes);
844 		break;
845 
846 		/* MMIO Read w/ sign-extension */
847 	case 0xbe:
848 		bytes = 1;
849 		fallthrough;
850 	case 0xbf:
851 		if (!bytes)
852 			bytes = 2;
853 
854 		ret = vc_do_mmio(ghcb, ctxt, bytes, true);
855 		if (ret)
856 			break;
857 
858 		/* Sign extend based on operand size */
859 		reg_data = vc_insn_get_reg(ctxt);
860 		if (!reg_data)
861 			return ES_DECODE_FAILED;
862 
863 		if (bytes == 1) {
864 			u8 *val = (u8 *)ghcb->shared_buffer;
865 
866 			sign_byte = (*val & 0x80) ? 0xff : 0x00;
867 		} else {
868 			u16 *val = (u16 *)ghcb->shared_buffer;
869 
870 			sign_byte = (*val & 0x8000) ? 0xff : 0x00;
871 		}
872 		memset(reg_data, sign_byte, insn->opnd_bytes);
873 
874 		memcpy(reg_data, ghcb->shared_buffer, bytes);
875 		break;
876 
877 	default:
878 		ret = ES_UNSUPPORTED;
879 	}
880 
881 	return ret;
882 }
883 
884 /*
885  * The MOVS instruction has two memory operands, which raises the
886  * problem that it is not known whether the access to the source or the
887  * destination caused the #VC exception (and hence whether an MMIO read
888  * or write operation needs to be emulated).
889  *
890  * Instead of playing games with walking page-tables and trying to guess
891  * whether the source or destination is an MMIO range, split the move
892  * into two operations, a read and a write with only one memory operand.
893  * This will cause a nested #VC exception on the MMIO address which can
894  * then be handled.
895  *
896  * This implementation has the benefit that it also supports MOVS where
897  * source _and_ destination are MMIO regions.
898  *
899  * It will slow MOVS on MMIO down a lot, but in SEV-ES guests it is a
900  * rare operation. If it turns out to be a performance problem the split
901  * operations can be moved to memcpy_fromio() and memcpy_toio().
902  */
903 static enum es_result vc_handle_mmio_movs(struct es_em_ctxt *ctxt,
904 					  unsigned int bytes)
905 {
906 	unsigned long ds_base, es_base;
907 	unsigned char *src, *dst;
908 	unsigned char buffer[8];
909 	enum es_result ret;
910 	bool rep;
911 	int off;
912 
913 	ds_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_DS);
914 	es_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_ES);
915 
916 	if (ds_base == -1L || es_base == -1L) {
917 		ctxt->fi.vector = X86_TRAP_GP;
918 		ctxt->fi.error_code = 0;
919 		return ES_EXCEPTION;
920 	}
921 
922 	src = ds_base + (unsigned char *)ctxt->regs->si;
923 	dst = es_base + (unsigned char *)ctxt->regs->di;
924 
925 	ret = vc_read_mem(ctxt, src, buffer, bytes);
926 	if (ret != ES_OK)
927 		return ret;
928 
929 	ret = vc_write_mem(ctxt, dst, buffer, bytes);
930 	if (ret != ES_OK)
931 		return ret;
932 
933 	if (ctxt->regs->flags & X86_EFLAGS_DF)
934 		off = -bytes;
935 	else
936 		off =  bytes;
937 
938 	ctxt->regs->si += off;
939 	ctxt->regs->di += off;
940 
941 	rep = insn_has_rep_prefix(&ctxt->insn);
942 	if (rep)
943 		ctxt->regs->cx -= 1;
944 
945 	if (!rep || ctxt->regs->cx == 0)
946 		return ES_OK;
947 	else
948 		return ES_RETRY;
949 }
950 
951 static enum es_result vc_handle_mmio(struct ghcb *ghcb,
952 				     struct es_em_ctxt *ctxt)
953 {
954 	struct insn *insn = &ctxt->insn;
955 	unsigned int bytes = 0;
956 	enum es_result ret;
957 	long *reg_data;
958 
959 	switch (insn->opcode.bytes[0]) {
960 	/* MMIO Write */
961 	case 0x88:
962 		bytes = 1;
963 		fallthrough;
964 	case 0x89:
965 		if (!bytes)
966 			bytes = insn->opnd_bytes;
967 
968 		reg_data = vc_insn_get_reg(ctxt);
969 		if (!reg_data)
970 			return ES_DECODE_FAILED;
971 
972 		memcpy(ghcb->shared_buffer, reg_data, bytes);
973 
974 		ret = vc_do_mmio(ghcb, ctxt, bytes, false);
975 		break;
976 
977 	case 0xc6:
978 		bytes = 1;
979 		fallthrough;
980 	case 0xc7:
981 		if (!bytes)
982 			bytes = insn->opnd_bytes;
983 
984 		memcpy(ghcb->shared_buffer, insn->immediate1.bytes, bytes);
985 
986 		ret = vc_do_mmio(ghcb, ctxt, bytes, false);
987 		break;
988 
989 		/* MMIO Read */
990 	case 0x8a:
991 		bytes = 1;
992 		fallthrough;
993 	case 0x8b:
994 		if (!bytes)
995 			bytes = insn->opnd_bytes;
996 
997 		ret = vc_do_mmio(ghcb, ctxt, bytes, true);
998 		if (ret)
999 			break;
1000 
1001 		reg_data = vc_insn_get_reg(ctxt);
1002 		if (!reg_data)
1003 			return ES_DECODE_FAILED;
1004 
1005 		/* Zero-extend for 32-bit operation */
1006 		if (bytes == 4)
1007 			*reg_data = 0;
1008 
1009 		memcpy(reg_data, ghcb->shared_buffer, bytes);
1010 		break;
1011 
1012 		/* MOVS instruction */
1013 	case 0xa4:
1014 		bytes = 1;
1015 		fallthrough;
1016 	case 0xa5:
1017 		if (!bytes)
1018 			bytes = insn->opnd_bytes;
1019 
1020 		ret = vc_handle_mmio_movs(ctxt, bytes);
1021 		break;
1022 		/* Two-Byte Opcodes */
1023 	case 0x0f:
1024 		ret = vc_handle_mmio_twobyte_ops(ghcb, ctxt);
1025 		break;
1026 	default:
1027 		ret = ES_UNSUPPORTED;
1028 	}
1029 
1030 	return ret;
1031 }
1032 
1033 static enum es_result vc_handle_dr7_write(struct ghcb *ghcb,
1034 					  struct es_em_ctxt *ctxt)
1035 {
1036 	struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1037 	long val, *reg = vc_insn_get_rm(ctxt);
1038 	enum es_result ret;
1039 
1040 	if (!reg)
1041 		return ES_DECODE_FAILED;
1042 
1043 	val = *reg;
1044 
1045 	/* Upper 32 bits must be written as zeroes */
1046 	if (val >> 32) {
1047 		ctxt->fi.vector = X86_TRAP_GP;
1048 		ctxt->fi.error_code = 0;
1049 		return ES_EXCEPTION;
1050 	}
1051 
1052 	/* Clear out other reserved bits and set bit 10 */
1053 	val = (val & 0xffff23ffL) | BIT(10);
1054 
1055 	/* Early non-zero writes to DR7 are not supported */
1056 	if (!data && (val & ~DR7_RESET_VALUE))
1057 		return ES_UNSUPPORTED;
1058 
1059 	/* Using a value of 0 for ExitInfo1 means RAX holds the value */
1060 	ghcb_set_rax(ghcb, val);
1061 	ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WRITE_DR7, 0, 0);
1062 	if (ret != ES_OK)
1063 		return ret;
1064 
1065 	if (data)
1066 		data->dr7 = val;
1067 
1068 	return ES_OK;
1069 }
1070 
1071 static enum es_result vc_handle_dr7_read(struct ghcb *ghcb,
1072 					 struct es_em_ctxt *ctxt)
1073 {
1074 	struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1075 	long *reg = vc_insn_get_rm(ctxt);
1076 
1077 	if (!reg)
1078 		return ES_DECODE_FAILED;
1079 
1080 	if (data)
1081 		*reg = data->dr7;
1082 	else
1083 		*reg = DR7_RESET_VALUE;
1084 
1085 	return ES_OK;
1086 }
1087 
1088 static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,
1089 				       struct es_em_ctxt *ctxt)
1090 {
1091 	return sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WBINVD, 0, 0);
1092 }
1093 
1094 static enum es_result vc_handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
1095 {
1096 	enum es_result ret;
1097 
1098 	ghcb_set_rcx(ghcb, ctxt->regs->cx);
1099 
1100 	ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_RDPMC, 0, 0);
1101 	if (ret != ES_OK)
1102 		return ret;
1103 
1104 	if (!(ghcb_rax_is_valid(ghcb) && ghcb_rdx_is_valid(ghcb)))
1105 		return ES_VMM_ERROR;
1106 
1107 	ctxt->regs->ax = ghcb->save.rax;
1108 	ctxt->regs->dx = ghcb->save.rdx;
1109 
1110 	return ES_OK;
1111 }
1112 
1113 static enum es_result vc_handle_monitor(struct ghcb *ghcb,
1114 					struct es_em_ctxt *ctxt)
1115 {
1116 	/*
1117 	 * Treat it as a NOP and do not leak a physical address to the
1118 	 * hypervisor.
1119 	 */
1120 	return ES_OK;
1121 }
1122 
1123 static enum es_result vc_handle_mwait(struct ghcb *ghcb,
1124 				      struct es_em_ctxt *ctxt)
1125 {
1126 	/* Treat the same as MONITOR/MONITORX */
1127 	return ES_OK;
1128 }
1129 
1130 static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,
1131 					struct es_em_ctxt *ctxt)
1132 {
1133 	enum es_result ret;
1134 
1135 	ghcb_set_rax(ghcb, ctxt->regs->ax);
1136 	ghcb_set_cpl(ghcb, user_mode(ctxt->regs) ? 3 : 0);
1137 
1138 	if (x86_platform.hyper.sev_es_hcall_prepare)
1139 		x86_platform.hyper.sev_es_hcall_prepare(ghcb, ctxt->regs);
1140 
1141 	ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_VMMCALL, 0, 0);
1142 	if (ret != ES_OK)
1143 		return ret;
1144 
1145 	if (!ghcb_rax_is_valid(ghcb))
1146 		return ES_VMM_ERROR;
1147 
1148 	ctxt->regs->ax = ghcb->save.rax;
1149 
1150 	/*
1151 	 * Call sev_es_hcall_finish() after regs->ax is already set.
1152 	 * This allows the hypervisor handler to overwrite it again if
1153 	 * necessary.
1154 	 */
1155 	if (x86_platform.hyper.sev_es_hcall_finish &&
1156 	    !x86_platform.hyper.sev_es_hcall_finish(ghcb, ctxt->regs))
1157 		return ES_VMM_ERROR;
1158 
1159 	return ES_OK;
1160 }
1161 
1162 static enum es_result vc_handle_trap_ac(struct ghcb *ghcb,
1163 					struct es_em_ctxt *ctxt)
1164 {
1165 	/*
1166 	 * Calling ecx_alignment_check() directly does not work, because it
1167 	 * enables IRQs and the GHCB is active. Forward the exception and call
1168 	 * it later from vc_forward_exception().
1169 	 */
1170 	ctxt->fi.vector = X86_TRAP_AC;
1171 	ctxt->fi.error_code = 0;
1172 	return ES_EXCEPTION;
1173 }
1174 
1175 static __always_inline void vc_handle_trap_db(struct pt_regs *regs)
1176 {
1177 	if (user_mode(regs))
1178 		noist_exc_debug(regs);
1179 	else
1180 		exc_debug(regs);
1181 }
1182 
1183 static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,
1184 					 struct ghcb *ghcb,
1185 					 unsigned long exit_code)
1186 {
1187 	enum es_result result;
1188 
1189 	switch (exit_code) {
1190 	case SVM_EXIT_READ_DR7:
1191 		result = vc_handle_dr7_read(ghcb, ctxt);
1192 		break;
1193 	case SVM_EXIT_WRITE_DR7:
1194 		result = vc_handle_dr7_write(ghcb, ctxt);
1195 		break;
1196 	case SVM_EXIT_EXCP_BASE + X86_TRAP_AC:
1197 		result = vc_handle_trap_ac(ghcb, ctxt);
1198 		break;
1199 	case SVM_EXIT_RDTSC:
1200 	case SVM_EXIT_RDTSCP:
1201 		result = vc_handle_rdtsc(ghcb, ctxt, exit_code);
1202 		break;
1203 	case SVM_EXIT_RDPMC:
1204 		result = vc_handle_rdpmc(ghcb, ctxt);
1205 		break;
1206 	case SVM_EXIT_INVD:
1207 		pr_err_ratelimited("#VC exception for INVD??? Seriously???\n");
1208 		result = ES_UNSUPPORTED;
1209 		break;
1210 	case SVM_EXIT_CPUID:
1211 		result = vc_handle_cpuid(ghcb, ctxt);
1212 		break;
1213 	case SVM_EXIT_IOIO:
1214 		result = vc_handle_ioio(ghcb, ctxt);
1215 		break;
1216 	case SVM_EXIT_MSR:
1217 		result = vc_handle_msr(ghcb, ctxt);
1218 		break;
1219 	case SVM_EXIT_VMMCALL:
1220 		result = vc_handle_vmmcall(ghcb, ctxt);
1221 		break;
1222 	case SVM_EXIT_WBINVD:
1223 		result = vc_handle_wbinvd(ghcb, ctxt);
1224 		break;
1225 	case SVM_EXIT_MONITOR:
1226 		result = vc_handle_monitor(ghcb, ctxt);
1227 		break;
1228 	case SVM_EXIT_MWAIT:
1229 		result = vc_handle_mwait(ghcb, ctxt);
1230 		break;
1231 	case SVM_EXIT_NPF:
1232 		result = vc_handle_mmio(ghcb, ctxt);
1233 		break;
1234 	default:
1235 		/*
1236 		 * Unexpected #VC exception
1237 		 */
1238 		result = ES_UNSUPPORTED;
1239 	}
1240 
1241 	return result;
1242 }
1243 
1244 static __always_inline void vc_forward_exception(struct es_em_ctxt *ctxt)
1245 {
1246 	long error_code = ctxt->fi.error_code;
1247 	int trapnr = ctxt->fi.vector;
1248 
1249 	ctxt->regs->orig_ax = ctxt->fi.error_code;
1250 
1251 	switch (trapnr) {
1252 	case X86_TRAP_GP:
1253 		exc_general_protection(ctxt->regs, error_code);
1254 		break;
1255 	case X86_TRAP_UD:
1256 		exc_invalid_op(ctxt->regs);
1257 		break;
1258 	case X86_TRAP_AC:
1259 		exc_alignment_check(ctxt->regs, error_code);
1260 		break;
1261 	default:
1262 		pr_emerg("Unsupported exception in #VC instruction emulation - can't continue\n");
1263 		BUG();
1264 	}
1265 }
1266 
1267 static __always_inline bool on_vc_fallback_stack(struct pt_regs *regs)
1268 {
1269 	unsigned long sp = (unsigned long)regs;
1270 
1271 	return (sp >= __this_cpu_ist_bottom_va(VC2) && sp < __this_cpu_ist_top_va(VC2));
1272 }
1273 
1274 /*
1275  * Main #VC exception handler. It is called when the entry code was able to
1276  * switch off the IST to a safe kernel stack.
1277  *
1278  * With the current implementation it is always possible to switch to a safe
1279  * stack because #VC exceptions only happen at known places, like intercepted
1280  * instructions or accesses to MMIO areas/IO ports. They can also happen with
1281  * code instrumentation when the hypervisor intercepts #DB, but the critical
1282  * paths are forbidden to be instrumented, so #DB exceptions currently also
1283  * only happen in safe places.
1284  */
1285 DEFINE_IDTENTRY_VC_SAFE_STACK(exc_vmm_communication)
1286 {
1287 	struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1288 	irqentry_state_t irq_state;
1289 	struct ghcb_state state;
1290 	struct es_em_ctxt ctxt;
1291 	enum es_result result;
1292 	struct ghcb *ghcb;
1293 
1294 	/*
1295 	 * Handle #DB before calling into !noinstr code to avoid recursive #DB.
1296 	 */
1297 	if (error_code == SVM_EXIT_EXCP_BASE + X86_TRAP_DB) {
1298 		vc_handle_trap_db(regs);
1299 		return;
1300 	}
1301 
1302 	irq_state = irqentry_nmi_enter(regs);
1303 	lockdep_assert_irqs_disabled();
1304 	instrumentation_begin();
1305 
1306 	/*
1307 	 * This is invoked through an interrupt gate, so IRQs are disabled. The
1308 	 * code below might walk page-tables for user or kernel addresses, so
1309 	 * keep the IRQs disabled to protect us against concurrent TLB flushes.
1310 	 */
1311 
1312 	ghcb = sev_es_get_ghcb(&state);
1313 	if (!ghcb) {
1314 		/*
1315 		 * Mark GHCBs inactive so that panic() is able to print the
1316 		 * message.
1317 		 */
1318 		data->ghcb_active        = false;
1319 		data->backup_ghcb_active = false;
1320 
1321 		panic("Unable to handle #VC exception! GHCB and Backup GHCB are already in use");
1322 	}
1323 
1324 	vc_ghcb_invalidate(ghcb);
1325 	result = vc_init_em_ctxt(&ctxt, regs, error_code);
1326 
1327 	if (result == ES_OK)
1328 		result = vc_handle_exitcode(&ctxt, ghcb, error_code);
1329 
1330 	sev_es_put_ghcb(&state);
1331 
1332 	/* Done - now check the result */
1333 	switch (result) {
1334 	case ES_OK:
1335 		vc_finish_insn(&ctxt);
1336 		break;
1337 	case ES_UNSUPPORTED:
1338 		pr_err_ratelimited("Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1339 				   error_code, regs->ip);
1340 		goto fail;
1341 	case ES_VMM_ERROR:
1342 		pr_err_ratelimited("Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1343 				   error_code, regs->ip);
1344 		goto fail;
1345 	case ES_DECODE_FAILED:
1346 		pr_err_ratelimited("Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1347 				   error_code, regs->ip);
1348 		goto fail;
1349 	case ES_EXCEPTION:
1350 		vc_forward_exception(&ctxt);
1351 		break;
1352 	case ES_RETRY:
1353 		/* Nothing to do */
1354 		break;
1355 	default:
1356 		pr_emerg("Unknown result in %s():%d\n", __func__, result);
1357 		/*
1358 		 * Emulating the instruction which caused the #VC exception
1359 		 * failed - can't continue so print debug information
1360 		 */
1361 		BUG();
1362 	}
1363 
1364 out:
1365 	instrumentation_end();
1366 	irqentry_nmi_exit(regs, irq_state);
1367 
1368 	return;
1369 
1370 fail:
1371 	if (user_mode(regs)) {
1372 		/*
1373 		 * Do not kill the machine if user-space triggered the
1374 		 * exception. Send SIGBUS instead and let user-space deal with
1375 		 * it.
1376 		 */
1377 		force_sig_fault(SIGBUS, BUS_OBJERR, (void __user *)0);
1378 	} else {
1379 		pr_emerg("PANIC: Unhandled #VC exception in kernel space (result=%d)\n",
1380 			 result);
1381 
1382 		/* Show some debug info */
1383 		show_regs(regs);
1384 
1385 		/* Ask hypervisor to sev_es_terminate */
1386 		sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1387 
1388 		/* If that fails and we get here - just panic */
1389 		panic("Returned from Terminate-Request to Hypervisor\n");
1390 	}
1391 
1392 	goto out;
1393 }
1394 
1395 /* This handler runs on the #VC fall-back stack. It can cause further #VC exceptions */
1396 DEFINE_IDTENTRY_VC_IST(exc_vmm_communication)
1397 {
1398 	instrumentation_begin();
1399 	panic("Can't handle #VC exception from unsupported context\n");
1400 	instrumentation_end();
1401 }
1402 
1403 DEFINE_IDTENTRY_VC(exc_vmm_communication)
1404 {
1405 	if (likely(!on_vc_fallback_stack(regs)))
1406 		safe_stack_exc_vmm_communication(regs, error_code);
1407 	else
1408 		ist_exc_vmm_communication(regs, error_code);
1409 }
1410 
1411 bool __init handle_vc_boot_ghcb(struct pt_regs *regs)
1412 {
1413 	unsigned long exit_code = regs->orig_ax;
1414 	struct es_em_ctxt ctxt;
1415 	enum es_result result;
1416 
1417 	/* Do initial setup or terminate the guest */
1418 	if (unlikely(boot_ghcb == NULL && !sev_es_setup_ghcb()))
1419 		sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1420 
1421 	vc_ghcb_invalidate(boot_ghcb);
1422 
1423 	result = vc_init_em_ctxt(&ctxt, regs, exit_code);
1424 	if (result == ES_OK)
1425 		result = vc_handle_exitcode(&ctxt, boot_ghcb, exit_code);
1426 
1427 	/* Done - now check the result */
1428 	switch (result) {
1429 	case ES_OK:
1430 		vc_finish_insn(&ctxt);
1431 		break;
1432 	case ES_UNSUPPORTED:
1433 		early_printk("PANIC: Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1434 				exit_code, regs->ip);
1435 		goto fail;
1436 	case ES_VMM_ERROR:
1437 		early_printk("PANIC: Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1438 				exit_code, regs->ip);
1439 		goto fail;
1440 	case ES_DECODE_FAILED:
1441 		early_printk("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1442 				exit_code, regs->ip);
1443 		goto fail;
1444 	case ES_EXCEPTION:
1445 		vc_early_forward_exception(&ctxt);
1446 		break;
1447 	case ES_RETRY:
1448 		/* Nothing to do */
1449 		break;
1450 	default:
1451 		BUG();
1452 	}
1453 
1454 	return true;
1455 
1456 fail:
1457 	show_regs(regs);
1458 
1459 	while (true)
1460 		halt();
1461 }
1462